Sparse model selection via integral terms

Area of Science:

• Computational Science
• Applied Mathematics
• Systems Biology

Background:

• Accurate dynamical system modeling is crucial for integrating experimental data, theory, and simulations.
• Identifying governing equations from data remains a significant challenge, especially with noise and numerous potential terms.
• Robust parameter estimation and model selection are vital for scientific discovery.

Purpose of the Study:

• To develop a novel learning-based approach for automated dynamical system identification directly from noisy data.
• To enable the selection of parsimonious models by identifying a minimal set of relevant features.
• To demonstrate the method's effectiveness across diverse complex systems.

Main Methods:

• Utilized a nonconvex sparse regression model to extract a small subset of important features from an overdetermined set.
• Developed a learning approach to fit noisy data to the trajectory of a dynamical system.
• Employed computational experiments to assess model stability, noise robustness, and recovery accuracy.

Main Results:

• The proposed sparse regression model effectively identifies dynamical systems from noisy experimental data.
• Demonstrated high stability, robustness to noise, and accurate recovery of system parameters.
• Successfully applied the method to various systems, including nonlinear equations, population dynamics, chaotic systems, and fast-slow systems.

Conclusions:

• The learning approach provides an effective and robust method for dynamical system identification and model selection.
• The technique facilitates the discovery of parsimonious and accurate models from complex, noisy datasets.
• This work advances the integration of data-driven methods with scientific theory for precise simulations.